ai-lsc/src/ai_lsc/registry/stack_templates/open-webui-full-rag.json

29 lines
1.9 KiB
JSON

{
"id": "open-webui-full-rag",
"name": "Open WebUI — Full RAG Knowledge Stack",
"description": "The most popular self-hosted ChatGPT replacement. Open WebUI + Ollama + ChromaDB + Whisper for voice input + Docling for document ingestion. The stack thousands of homelabbers and YouTubers run for private AI chat with full document understanding.",
"version": "1.0",
"author": "ai-lsc",
"tags": ["open-webui", "chatgpt-alternative", "rag", "homelab", "youtube-trending", "document-ai", "local-first"],
"endpoints": {
"ollama_base": "http://localhost:11434",
"openwebui": "http://localhost:3000",
"chromadb": "http://localhost:8000"
},
"tools": [
"ollama",
"openwebui",
"chromadb",
"whisper",
"docling",
"markitdown"
],
"notes": {
"youtube_context": "The #1 homelab AI setup across YouTube. Every self-hosted AI tutorial covers this exact stack. Channels like @techhut, @crosstalksolutions, and @networkchuck have dedicated videos with 500K+ views on this combo.",
"recommended_models": "llama3.1:8b (fast chat), llama3.1:70b (quality), qwen2.5:32b (multilingual), mistral-nemo:12b (lightweight), nomic-embed-text (embeddings)",
"setup": "Open WebUI auto-detects Ollama on localhost:11434. Upload PDFs/docs in the UI — they get chunked and embedded into ChromaDB automatically. Whisper enables the microphone button for voice-to-text input.",
"workflow": "Documents → Docling/MarkItDown → Markdown → Open WebUI RAG pipeline → ChromaDB vector store. Whisper handles voice queries. All inference through Ollama on local GPU.",
"tips": "Set OLLAMA_NUM_PARALLEL=4 for concurrent chat requests. Use Open WebUI's built-in model switching to route simple queries to 8B and complex ones to 70B. Create workspaces per project for isolated document collections."
}
}